What Second Life could have been

[ChinRey]

I know I'm going to get a few likes for this post and I appreciate that a lot but I appreciate comments even more. I do not want this to be yet another "Rey showing off" post - I never actually wanted those, I was always looking for genuine discussions about how to make SL a better place for more people. Maybe it works this time.

Oh and if any of the moderators reads this, please don't move it to the cotnent creators forum. I posted this in the general section because I think this is something that really should concern everybody in SL, not only content creators. With that being said:

---

This is in a way a follow up to three posts I wrote back in July (starting here) where I pointed out just how inefficient SL content is and how much we could improve the build quality with better optimization. I got a bit of flame from people who don't know nearly as much about 3D modelling as they believe they do and I couldn't prove my point at that time. I can now though. No, nobody has offered to lend me the sims I need in SL but a gentleman named Michael Timeless has on Kitely so I made a forest. It's not quite finished and probably will never be (most of the trees are in place but most aren't animated yet and I've barely started on the undergrowth and such) and I'm running out of time so I have to post about it now. It's more than presentable as it is and certainly enough to prove the point I had. I really hope those of you who are on opensim too take the time to visit. The region is, as I said, on Kitely and it is called Fallingwater (no prize for guessing what kind of build it originally had). For everybody else, some pictures and data below.

---

But first, there's something else I'd also like to discuss. This is a build that is impossible in Second Life for a number of reasons, not only because of the huge tier cost. Realistically I think it is also impossible as a permanent build on opensim. But it could have been possible if SL had evolved along another path. I'm not saying SL should have taken that different path (or any of the other ones that were once possible). It is what it is and lots of people appreciate that, so that's fine. But even so, I can't help thinking what if...

Me, I certainly would have loved to have a virtual reality in the style this build only can hint at. A big, continuous but ever changing landscape you could immerse yourself in and where you could walk or drive for miles without interruptions. But maybe that's just me.

---

Anyways, here are the info and pics for those who can't go and see the (un)real thing.

The entire forest fills up a region 1024x1024 m in size, that's the equivalent of 16 Second Life size regions. The render load is low enough even my li'l off-the-shelf home computer with abysmal graphics performance and an oversized screen can easily handle utlra graphics and 1024 m draw distance. There are no collapsed LoD models. With LoD factor set to 1 you may notice occasional switches between LoD models but very rarely and only if you're really looking for them. With LoD factor 2 you will not see any LoD issues whatsoever.

As is is now, the build has

• 7,842 trees
• A fairly elaborate road system (this is a Chin Rey build after all)
• 443 meshes and one prim
• About 100-150 unique meshes
• 24 textures, all of them 512x512
• No scripts
• 142,479 tris

In Second Life the land impact would have been somewhere between 3,000-5,000. That sounds like a lot but remember, this is for 16 sims, not just one.

My budget for the finished build (if it ever is finished) was all in fives:

• 5 scripts
• 50 textures
• 500 unique meshes
• 5,000 meshes
• 500,000 tris

I'd probably end up with a few more scripts and textures but not nearly that many meshes and tris.

Edited October 6, 2020 by ChinRey
Typos

[PermaRuthed]

I could see myself going for a Sunday drive through this forest.

[ChinRey]

11 minutes ago, PermaRuthed said:

I could see myself going for a Sunday drive through this forest.

You can't on this demo unfortunately. Michael forgot to give me admin powers for the parcel, I only have it for the region. That means I can't open it up for people to rez. I should probably ask him to change that but I dont' know how long I have the sim. Originally I was onyl going to have it for two weeks but then he changed his mind and said a month. Three weeks have past now and although I have a strong feeling it'll stay up for a little bit longer, I may have to take it down early next week and if so, there's no point in making any changes.

But you get one of the main ideas, except it shouldn't stop there. That road shouldn't just go round and round through a forest on a small island. It should lead across a bridge to mainland, over to a main road, through farmland, villages, towns and wilderness, northwards to the subarctic, southwards to warmer climates or upwards into the mountains.

The total nominal areal of all regions in Second Life combined is about the same as London. In reality it's less than half that size because of the scale issues. That's not big enough for something like this and to make matters worse, it's fragmented. It's not a virtual world at all but rather a heap of tiny little scenarios only very loosely connected. This fragmentation was not Linden Lab's original intention, it just happened by accident but who knows, maybe it's what most people wanted. In any case, there's nothing that can eb done about it now.

[Arielle Popstar]

18 minutes ago, ChinRey said:

You can't on this demo unfortunately. Michael forgot to give me admin powers for the parcel, I only have it for the region. That means I can't open it up for people to rez.

Ask for an oar and it can be loaded anywhere on Opensim in the future.

[ChinRey]

2 hours ago, Arielle Popstar said:

Ask for an oar and it can be loaded anywhere on Opensim in the future.

Yes but it's just as well that its days are numbered. This is a demo, remember. If people don't see or don't want to see the point now, they never will.

I will keep an oar of course but I have no need to keep it online for my own sake and I have lots of other projects. If I ever reload it permanently somewhere it'll probably be as a residental region. It's deliberately made to be easily converted to hold up to 80 houses and one or two two small ports.

Edited October 6, 2020 by ChinRey

[Akane Nacht]

3 hours ago, ChinRey said:

Me, I certainly would have loved to have a virtual reality in the style this build only can hint at. A big, continuous but ever changing landscape you could immerse yourself in and where you could walk or drive for miles without interruptions. But maybe that's just me.

This kind of thing you describe already exists in open world games. For non builders like me, we can find rolling hills and grand forests to gallop across seamlessly already, at a tiny cost.

I think, visually, SL's strengths are in the details of small spaces, and avatar customisation of course. Large sweeping landscapes.. maybe not so much. I definitely see more people shopping or holed up in private homes than hanging around in big parks. I don't know of any other game where you can find the perfect sprig of orchids for your sideboard or the exact shade of butterfly to hover over your little wildflower patch.

I suppose there is no reason why we can't aim for both tho. 

Edited October 6, 2020 by Akane Nacht
typo

[ChinRey]

1 hour ago, Akane Nacht said:

This kind of thing you describe already exists in open world games. For non builders like me, we can find rolling hills and grand forests to gallop across seamlessly already, at a tiny cost.

Not tiny if you also count the hardware costs.

But the main problem with those are that they are monocultures. Each one has one single theme. It's the opposite problem of SL and it's just as significant. That's why I'm not too concerned about keeping my forest online permanently btw. On its own it's only another monoculture, allbeit with far lower hardware requirements. It's only if it was used as a part of a much larger unified context it would really make sense. This would require several content creators, not only because it's way too big a job for one, but even more because it needs people who see things from different perspectives, have different ideas and come from different cultural backgrounds. Yet they all have to be wiling and able to adapt their styles to fit the whole. That's a huge challenge even if the environment was created to encourage cooperation which SL and opensim most definitely aren't.

Btw, to get to the first point in my original post too, I don't think it's quite fair to compare my forest to high end game engine based environments because of the very different hardware requirements. My aim was to ceate an visual quality comparable to SL only with much longer draw distance and only a fraction of the server and client load. That's easy enough and I honestly don't understand why nobody else does it.

 

1 hour ago, Akane Nacht said:

I think, visually, SL's strengths are in the details of small spaces, and avatar customisation of course.

Oh yes, definitely. That is the path SL ended up taking.

 

1 hour ago, Akane Nacht said:

I suppose there is no reason why we can't aim for both tho. 

Oh, there are lots of reasons. I already mentioned the tier. Perhaps I should tell you why Michael don't need his Kitely region himself at the moment. It's because he's working on a different project elsewhere, far more interesting than my little forest. It's a 16x16 (that's 256 SL regions!) reproduction of a national park, specially made as an experience for disabled vets who are on longer able to get out in the nature in RL. Imagine how much it would cost to host something like that in SL and it's still fairly small, only about 16 km2. The high tier in SL is of course much because LL charges top dollars for their hosting but it's just as much because they assign such a lavish amount of resources to each region. I don't think I've ever seen a region that really needed the 5,000 LI worth of resources and certainly not 25,000. But you get it anyway and you pay for it anyway so you might as well use it and so there's no real incentive to try to build with efficient content.

Other reasons why a Big Immersive World concept wont' work in SL incude walking speed, camera position, the complexity of the messy and bloated UI etc., etc. And I can't see any reason why anybody would want to try to make something like this in SL at this stage. Practically everything would have to be made from scratch anyway and the people such a world would appeal to are generally not SL users. Opensim would have been much cheaper and just as good but still with many of SL's limitations. That's why I called this thread what SL could have been, not what it can become. It's not what it should be either - I don't have the answer to that.

Edited October 6, 2020 by ChinRey

[Paul Hexem]

Every time a thread like this comes up, I think about some of the modern survival games out there.

Huge sprawling maps under constant terraforming and deforestation/reforestation, people building structures in real time in the game, custom textures uploaded, dozens of players running around, AI wandering around... no matter what FPS you're getting, you can still run around, engage in combat, and get things done without lagging or rubber banding. But then in SL, a region can grind to a halt because 20 people are just standing there. On a 300 USD/month server, vs the survival's 15 USD a month server.

Its easy to talk about how assets are delivered being so different... But thanks to local caching, that difference should only last long enough to download what's in the region. After that, there's no excuse for the amount of server lag we deal with, not for the prices we pay.

[animats]

17 minutes ago, Paul Hexem said:

Huge sprawling maps under constant terraforming and deforestation/reforestation, people building structures in real time in the game, custom textures uploaded, dozens of players running around, AI wandering around... no matter what FPS you're getting, you can still run around, engage in combat, and get things done without lagging or rubber banding. But then in SL, a region can grind to a halt because 20 people are just standing there.

Yes. There are too many big user-modifiable worlds out there now. It's no longer credible to claim that it's impossible to have a high-performing Second Life. Yes, it's hard. But not impossible.

Many of SL's performance problems are simply bugs:

• About half of simulator time goes down the drain because scripts that are idle still use CPU time. Not a lot per script, but enough that 5000 to 6000 scripts just sitting there uses 100% of script time for a full region.
• Region crossing failures. Years of work, and they're still not reliable. Part of this is a protocol problem. The avatar's viewer is involved in one handshake that holds up the whole region crossing. And double region crossings, at corners, are simply broken. The system cannot handle entering as second region before the first region crossing completes. Even in "the cloud". (Hit a corner in that test Blake Sea region within 1m of the corner at moderate speed, and you're out. Corners in the middle of roads are not rare. Kama City has one on every intersection.) But nothing prevents a double region crossing.
• There's a huge load on a sim as as an avatar enters. Watch the "nearby users" window while walking in a crowded sim. Watch walking stall when a new avatar appears.

Fix those three, and SL sim-side would work much better. They're all hard problems, but those are the big three.

Viewer side, there's a longer list. But the viewer isn't that different from a modern big-world MMO client where content comes in from the network.

 

[ChinRey]

2 hours ago, animats said:

Fix those three, and SL sim-side would work much better. They're all hard problems, but those are the big three.

I'm sure you are right and I could add a few things to the list of software issues slowing SL down. But you are make a huge mistake if you blame it all on the software. GIGO still applies and asset efficiency is at least as imortant for performance. I had a visitor while I was pretending to work on the forest earleir today. She told me she typically got about 40 fps at regions comparable to the forest. At the forest she had a consistent 70. That's a 75% improvement - less than I had hoped for and slightly less than I get myself but still quite significant. This gain is not from reducing the visual quality nor is it from improved software, ti's all about using meshes and textures that make the most out of the render engine we have at the moment.

Another interesting and unexpected stat from that forest. One peculiarity of Kitely is that it takes regions offline when they are empty and restart them when somebody tries to enter. That means we can learn something about server load time too. Typically a fully developed 4x4 takes two or three minutes to launch - I've seen regions that take more than five minutes. My forest launches in less than 10 seconds.

Just like the software, asset performance is also to a large degree LL's responsibility. They haven't done anything worth speaking of to encgourage performant content or even ifnorm people how to do it and the performance of content they provide themselves these days is abysmal. But uike the software issues it's still something everybody in SL can do something about. If they want to.

(Edit: For those not famiiar with GIGO, it's old computerish and means "Garbage In, Garbage Out".)

Edited October 6, 2020 by ChinRey

[animats]

4 hours ago, ChinRey said:

But you are make a huge mistake if you blame it all on the software. GIGO still applies and asset efficiency is at least as imortant for performance.

That reflects SL's rather weak level of detail system. It's fine to have high detail when you're really close. That's one of SL's great strengths. We see that on most of Strawberry Linden's posts of cluttered rooms.

But below "High" level of detail, most of the levels of detail range from poor to terrible. Chin puts effort into lower levels of detail. Most creators don't. Go to stores where non-clothing mesh items are on sale, and  set your LOD factor to 0. It looks awful.

Level of detail has to be automated somehow. That's common for games today. Sansar had it. UE5 has a new system for that, as can be seen in their demo, and the first release of UE5 is supposed to be out next year. I've tried some schemes for generating impostors for SL. There are ways to deal with this problem. It's hard, but possible.

(Clothing is a special case. None of the mesh reduction algorithms out there do a good job on thin objects with folds in them, which is SL clothing. SL already has avatar impostors, though.)

[ChinRey]

9 hours ago, animats said:

That reflects SL's rather weak level of detail system.

They say where ther's a will there's a way. But in this case there is no will.

 

9 hours ago, animats said:

It's fine to have high detail when you're really close. That's one of SL's great strengths. We see that on most of Strawberry Linden's posts of cluttered rooms.

There is no limit to how high details you can have really close if there is no limit to how much computing powers a single object is allowed to hog. That's true for all engines, nothing special about Second Life there. The difference is that all other engines have mechanisms to restrict this. They may not be formalized and certainly aren't blatantly obvious but they are still very much there. Second Life has no such limtiations. Port a Strawberry cluttered room or any other SL scene for that matter to UE4 or Unity with all the tris and vertices and pixels and it'll be just as laggy there as it is in SL.

Many of us remember the spectacular 2014 UE4 demo and we all went ooh and aah over it. I did myself as late as yesterday. But did we ever really look at it with content creator eyes? And did we ever reflect on how quickly people were able to create similar scenes in Unity with no software upgrades.

Look at it again. Notice how simple verything actually is. I wouldn't be surprised at all if that entire scene uses less tris and pixels than an average SL avatar. Yet it looks so realistic (apart from the obvious health and safety issues but that's a completely different matter.) The effect didn't actually have anything to do with the big UE upgrade to version 4. It was all about efficient content and PBR. We don't have PBR in SL and that may well be too much to ask for. But we could have had efficient content. It's too late now of course but I still can't help thinking of what Second Life could have been.

 

Edit: Forgot to post the video. Several people made walkthrough videos but I think this one is aprticularly itneresting:

 

Edited October 7, 2020 by ChinRey

[Profaitchikenz Haiku]

I fear LL are in between Scylla and Charybdis here. Some people wander around viewing a region with a moderate draw distance, some people like to stand at a landing point and cam around the entire region without moving. I cannot see a way to optimise LoD levels to fit both those types of behaviour.

I know it's been disparaged by many. but playing around with RenderVolumeLodFactor is a perfectly good workaround, moderate draw distance users can get by with the LL-recommended setting, hard-core 256metre+ draw-distance users can crank the figure all the way up to 11.

Another group of users are the photographers and film-makers. They are probably the ones more affected by this because medium to long shots which are desirable to avoid the excessive perspective leaning are going to experience the degeneration, and in the case of film-makers, cranking up the setting is possibly going to hit their frame rates.

The vehicle-users possibly get hit the worst as they are travelling fast enough to need rapid redrawing but cannot ramp up the LodFactor as that is going to get them worse frame rates. I can't see any workable medium term solution to this other than avoiding regions that are hard to view, and persuading users, as Chin is trying, to adopt better methods.

I'm impressed buy Chin's original post, it reminds me of the tricks we used in the train simulators to try and stop the annoying objects popping into view full-blown behaviour: judicious view-blocks combined with carefully crafted lower-LoD models worked.

Personally, I'm in favour of people walking or travelling around a region instead of just camming from a central point, but it's all about personal freedom.

[ChinRey]

 

3 hours ago, Profaitchikenz Haiku said:

I'm impressed buy Chin's original post, it reminds me of the tricks we used in the train simulators to try and stop the annoying objects popping into view full-blown behaviour: judicious view-blocks combined with carefully crafted lower-LoD models worked.

Thank you but I do think you are underestimating my forest a bit. I avoid annoying objects popping into view simply by not having any in the region. There is no need to block anything from being seen at any distance (except for one pile of rock, not sure if I should do anything about that).

3 hours ago, Profaitchikenz Haiku said:

Personally, I'm in favour of people walking or travelling around a region instead of just camming from a central point, but it's all about personal freedom.

Me too and that's what my forest is for. These are not background trees in any way. Each and every one of the 8.010 (yes, I've added a few since my first post) trees have a mesh trunk and a foliage that is a bit more than just two crossed sheets. Some of them turned out to be a bit too simple and I was also a bit too cautious about physics so some of the trunks are phantoma t the moment. But I have plenty of room on my resource budget to fix those issues and even as they are, you should still beable to enjoy a walk around them. I have of course placed some more elaborate ones in strategic places but even the bulk of the forest is what I call "playground content", that is if you are among them and involved in some activity such as role playing, playing a game, driving or walking around or chatting, you shouldn't anything wrong. If you do, it's a bug.

If I were to make a background forest that was only supposed to be seen from the outside, it would have looked a bit like this (quick scaled down mockup in a sandbox) - 2,680 trees, 5,376 tris, three 512x512 textures:

 

 

Edited October 7, 2020 by ChinRey

[Profaitchikenz Haiku]

6 minutes ago, ChinRey said:

I avoid annoying objects popping into view simply by not having any annoying objects in the region

OK, what I was referring to was the people who cannot create all the content themselves but mix in some of their own and some of other people's creations. In the train-sims, for example, you might make a few buildings yourself but find some available stations all ready to use, which however didn't have the best Lod behaviour, so you ended up putting perhaps trees in front of them along the longest direction of approach so as to occlude the object partially. Likewise in SecondLife., people tend to buy houses from other creators.

 

Trees in SL can be awful, but the best ones I saw were sculpts with the trunk and branches as one sculpt and the foliage as another sculpt comprisng several panes. Because of the way sculpts work they could be a bit messy during the intial loading phase but after that behaved perfectly at varying distances, and, of course, were just 2 LI.

 

You mentioned another point that I too have experienced, none of the regions I have visited in the Opensim have ever crashed my viewer or slogged it to a halt, but many do so in SecondLife. is it the prevalence of mesh with large textures and the extra bump and speculars for ALM? I don't know.

[ChinRey]

2 hours ago, Profaitchikenz Haiku said:

...

so you ended up putting perhaps trees in front of them along the longest direction of approach so as to occlude the object partially.

Yes, and that is very much a good practice in many ways. Ideally you don't want an object to have more elaborate LoD models than it actually needs. But to achieve that you need to think context first, content second. Each object has to be made specifically for a specific location in the scene. It limits the possibiities for reusing assets (a very important technique for reducing the load) though, it doesn't work with viewer adjustable LoD factor and such customisation is not really an option in SL anyway.

 

2 hours ago, Profaitchikenz Haiku said:

Trees in SL can be awful, but the best ones I saw were sculpts with the trunk and branches as one sculpt and the foliage as another sculpt comprisng several panes. Because of the way sculpts work they could be a bit messy during the intial loading phase but after that behaved perfectly at varying distances, and, of course, were just 2 LI.

A sculpt tree still has 4096 tris though and that is way too much for a single tree, unless it's a very elaborate feature item. Even so, it's not uncommon for mesh trees to waste even more tris so yes, you have a point. I once wrote a post at the OPQ blog, comparing one of my own trees to one of the Mole trees from Bellisseria. These are medium level trees, good enough to be up front but not intended as features. Even so, that Mole tree has a whopping 1080 tris! That's not at all unusual in SL, in fact it's probably lower than average and compared to that kind of meshes sculpts perform very well indeed. The best tree makers in SL and on opensim, however, people like Alex Bader, Reid Parkins, Ozwell Wayfarer, Eldowyn Inshan and Teresa Matfield, can easily run rings around that field without raising a sweat. The very best when it comes to high performance vegetation is as far ahead of them as they are ahead of the field but that's me so I probably should shut up now and besides, I never have time to actually list my plants for sale, I'm always way too busy procrastinating on the forums.

 

2 hours ago, Profaitchikenz Haiku said:

You mentioned another point that I too have experienced, none of the regions I have visited in the Opensim have ever crashed my viewer or slogged it to a halt, but many do so in SecondLife.

I've sertainly come across uber-laggy regions on opensim too but yes, they aren't nearly as common as they are in Second Life. I think there are several reasons for that.

One is that opensim isn't nearly as "meshified" as SL yet. It's still much about prims and you'd be hard put if you want to lag down a region with prims. The prim system was developed by Avi Bar-Zeev with a lot of help from Cory Ondrejka and Eric Call and unlike their successors at the Lab, those guys were really, really lag concious. I've heard a story about Eric chewing out a poor aspiring builder for using a single 1024 on a house and I think that says it all. Prims can be slow to load, mainly because of a flaw in the assets database structure, but once they're in place they hardly ever cause any lag issues worth speaking of.

Another reason is that is your opensim region lags, there's nobody to blame but yourself. You can't complain to the host because you are the host and you can't blame the neighbor's laggy build because there is no neighbor. If you have performance issues, it's up to you to fix it, nobody else. Unfortunately, the solution is often to strip down the build, not to replace heavy content with better performing equivalents.

A third reason is that much of the opensim population consists of people who fled SL because of the lag and they are of course likely to be a bit more concious of the problem than those who remained here.

The fourth reason... this is where my post can go horribly wrong. I have to expect a lot of flame for this and might even be banned from the forum. But oh well:

Even when there is mesh on opensim, there's not that much of the high poly and/or heavily textured kind that makes up sot of the commercial mesh in SL. Now, don't get me wrong here, I absolutely love some of those builds myself. I can rez an Apple Fall or Trompe Loeill build and just and admire it - they're true works of art. But that kind of content is not compatible with high performance and will never ever be. Yes, you can overcome the lag issues by buying a better computer, by butchering the LoD or by dropping your draw distance but all those solution conflict with the Big World concept. This is the main reason why I'm saying a build like my forest is impossible in Second Life. Even if somebody funded a brand new continent specifically for big low lag landscapes far away from the rest of SL, the high lag cotnent would slowly but surely leak in. One such object won't make any difference after all. Two? No problem. Three? Sure! And so and and so on until you add the straw that breaks the camel's back.

This is also why I'm emphasizing that my forest is a glimpse of what SL could have been, not what it should have been. SL became what it is because LL wasn't paying attention. It certanly wasn't planned and I do not believe it was what most users wanted (it probably is now though since most of those who were looking for something else have left). But there obviously is a need and a market for what it has become and it's probably gone too far down that path to turn around anyway.

That being said there is one big difference between the examples I mentioned. Works by artists like Apple Fall and Cory Edo may come at a high cost in eprformance and you may or may not be willing and able to pay that price. But they do deliver, you get what you pay for. That Mole tree on the other hand, that's sheer waste. It's just a mediocre filler tree any half baked bulder could make at half the render cost and a skilled one could do at a quarter. There's no real excuse for that.

Edited October 7, 2020 by ChinRey